Convolutional Neural Networks

Project: Write an Algorithm for a Dog Identification App


In this notebook, some template code has already been provided for you, and you will need to implement additional functionality to successfully complete this project. You will not need to modify the included code beyond what is requested. Sections that begin with '(IMPLEMENTATION)' in the header indicate that the following block of code will require additional functionality which you must provide. Instructions will be provided for each section, and the specifics of the implementation are marked in the code block with a 'TODO' statement. Please be sure to read the instructions carefully!

Note: Once you have completed all of the code implementations, you need to finalize your work by exporting the Jupyter Notebook as an HTML document. Before exporting the notebook to html, all of the code cells need to have been run so that reviewers can see the final implementation and output. You can then export the notebook by using the menu above and navigating to File -> Download as -> HTML (.html). Include the finished document along with this notebook as your submission.

In addition to implementing code, there will be questions that you must answer which relate to the project and your implementation. Each section where you will answer a question is preceded by a 'Question X' header. Carefully read each question and provide thorough answers in the following text boxes that begin with 'Answer:'. Your project submission will be evaluated based on your answers to each of the questions and the implementation you provide.

Note: Code and Markdown cells can be executed using the Shift + Enter keyboard shortcut. Markdown cells can be edited by double-clicking the cell to enter edit mode.

The rubric contains optional "Stand Out Suggestions" for enhancing the project beyond the minimum requirements. If you decide to pursue the "Stand Out Suggestions", you should include the code in this Jupyter notebook.


Why We're Here

In this notebook, you will make the first steps towards developing an algorithm that could be used as part of a mobile or web app. At the end of this project, your code will accept any user-supplied image as input. If a dog is detected in the image, it will provide an estimate of the dog's breed. If a human is detected, it will provide an estimate of the dog breed that is most resembling. The image below displays potential sample output of your finished project (... but we expect that each student's algorithm will behave differently!).

Sample Dog Output

In this real-world setting, you will need to piece together a series of models to perform different tasks; for instance, the algorithm that detects humans in an image will be different from the CNN that infers dog breed. There are many points of possible failure, and no perfect algorithm exists. Your imperfect solution will nonetheless create a fun user experience!

The Road Ahead

We break the notebook into separate steps. Feel free to use the links below to navigate the notebook.

  • Step 0: Import Datasets
  • Step 1: Detect Humans
  • Step 2: Detect Dogs
  • Step 3: Create a CNN to Classify Dog Breeds (from Scratch)
  • Step 4: Create a CNN to Classify Dog Breeds (using Transfer Learning)
  • Step 5: Write your Algorithm
  • Step 6: Test Your Algorithm

Step 0: Import Datasets

Make sure that you've downloaded the required human and dog datasets:

Note: if you are using the Udacity workspace, you DO NOT need to re-download these - they can be found in the /data folder as noted in the cell below.

  • Download the dog dataset. Unzip the folder and place it in this project's home directory, at the location /dog_images.

  • Download the human dataset. Unzip the folder and place it in the home directory, at location /lfw.

Note: If you are using a Windows machine, you are encouraged to use 7zip to extract the folder.

In the code cell below, we save the file paths for both the human (LFW) dataset and dog dataset in the numpy arrays human_files and dog_files.

In [1]:
%matplotlib inline
%config InlineBackend.figure_format = 'retina'
import matplotlib.pyplot as plt

plt.style.use('seaborn-colorblind')
In [2]:
import numpy as np
from glob import glob

# load filenames for human and dog images
human_files = np.array(glob("data/lfw/*/*"))
dog_files = np.array(glob("data/dog_images/*/*/*"))

# print number of images in each dataset
print('There are %d total human images.' % len(human_files))
print('There are %d total dog images.' % len(dog_files))
There are 13233 total human images.
There are 8351 total dog images.

Step 1: Detect Humans

In this section, we use OpenCV's implementation of Haar feature-based cascade classifiers to detect human faces in images.

OpenCV provides many pre-trained face detectors, stored as XML files on github. We have downloaded one of these detectors and stored it in the haarcascades directory. In the next code cell, we demonstrate how to use this detector to find human faces in a sample image.

In [3]:
import cv2                
import matplotlib.pyplot as plt                        
%matplotlib inline                               

# extract pre-trained face detector
face_cascade = cv2.CascadeClassifier('haarcascades/haarcascade_frontalface_alt.xml')

# load color (BGR) image
img = cv2.imread(human_files[0])
# convert BGR image to grayscale
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# find faces in image
faces = face_cascade.detectMultiScale(gray)

# print number of faces detected in the image
print('Number of faces detected:', len(faces))

# get bounding box for each detected face
for (x,y,w,h) in faces:
    # add bounding box to color image
    cv2.rectangle(img,(x,y),(x+w,y+h),(255,0,0),2)
    
# convert BGR image to RGB for plotting
cv_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

# display the image, along with bounding box
plt.imshow(cv_rgb)
plt.show()
Number of faces detected: 1

Before using any of the face detectors, it is standard procedure to convert the images to grayscale. The detectMultiScale function executes the classifier stored in face_cascade and takes the grayscale image as a parameter.

In the above code, faces is a numpy array of detected faces, where each row corresponds to a detected face. Each detected face is a 1D array with four entries that specifies the bounding box of the detected face. The first two entries in the array (extracted in the above code as x and y) specify the horizontal and vertical positions of the top left corner of the bounding box. The last two entries in the array (extracted here as w and h) specify the width and height of the box.

Write a Human Face Detector

We can use this procedure to write a function that returns True if a human face is detected in an image and False otherwise. This function, aptly named face_detector, takes a string-valued file path to an image as input and appears in the code block below.

In [4]:
# returns "True" if face is detected in image stored at img_path
def face_detector(img_path):
    img = cv2.imread(img_path)
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    faces = face_cascade.detectMultiScale(gray)
    return len(faces) > 0

(IMPLEMENTATION) Assess the Human Face Detector

Question 1: Use the code cell below to test the performance of the face_detector function.

  • What percentage of the first 100 images in human_files have a detected human face?
  • What percentage of the first 100 images in dog_files have a detected human face?

Ideally, we would like 100% of human images with a detected face and 0% of dog images with a detected face. You will see that our algorithm falls short of this goal, but still gives acceptable performance. We extract the file paths for the first 100 images from each of the datasets and store them in the numpy arrays human_files_short and dog_files_short.

Answer: (You can print out your results and/or write your percentages in this cell)

In [5]:
import warnings
warnings.simplefilter('ignore')

from tqdm import tqdm, trange, tqdm_notebook
#from tqdm.notebook import trange

human_files_short = human_files[:100]
dog_files_short = dog_files[:100]

#-#-# Do NOT modify the code above this line. #-#-#

## TODO: Test the performance of the face_detector algorithm 
## on the images in human_files_short and dog_files_short.
human = 0
dog = 0
#for i in trange(100):
for i in tqdm_notebook(range(100)):
    if face_detector(human_files_short[i]) == True:
        human += 1
    if face_detector(dog_files_short[i]) == True:
        dog += 1
        
print('Human face percentage {}%'.format(human))
print('Dog face percentage {}%'.format(dog))
Human face percentage 98%
Dog face percentage 9%

We suggest the face detector from OpenCV as a potential way to detect human images in your algorithm, but you are free to explore other approaches, especially approaches that make use of deep learning :). Please use the code cell below to design and test your own face detection algorithm. If you decide to pursue this optional task, report performance on human_files_short and dog_files_short.

In [6]:
### (Optional) 
### TODO: Test performance of anotherface detection algorithm.
### Feel free to use as many code cells as needed.
face_cascade2 = cv2.CascadeClassifier('haarcascades/haarcascade_frontalface_alt2.xml')

def face_detector2(img_path):
    img = cv2.imread(img_path)
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    faces = face_cascade2.detectMultiScale(gray)
    return len(faces) > 0

human = 0
dog = 0
for i in tqdm_notebook(range(100)):
    if face_detector2(human_files_short[i]) == True:
        human += 1
    if face_detector2(dog_files_short[i]) == True:
        dog += 1
        
print('Human face percentage {}%'.format(human))
print('Dog face percentage {}%'.format(dog))
Human face percentage 99%
Dog face percentage 16%

Step 2: Detect Dogs

In this section, we use a pre-trained model to detect dogs in images.

Obtain Pre-trained VGG-16 Model

The code cell below downloads the VGG-16 model, along with weights that have been trained on ImageNet, a very large, very popular dataset used for image classification and other vision tasks. ImageNet contains over 10 million URLs, each linking to an image containing an object from one of 1000 categories.

In [7]:
import torch
import torchvision.models as models

# define VGG16 model
VGG16 = models.vgg16(pretrained=True)

# check if CUDA is available
use_cuda = torch.cuda.is_available()

# move model to GPU if CUDA is available
if use_cuda:
    VGG16 = VGG16.cuda()
    

Given an image, this pre-trained VGG-16 model returns a prediction (derived from the 1000 possible categories in ImageNet) for the object that is contained in the image.

(IMPLEMENTATION) Making Predictions with a Pre-trained Model

In the next code cell, you will write a function that accepts a path to an image (such as 'dogImages/train/001.Affenpinscher/Affenpinscher_00001.jpg') as input and returns the index corresponding to the ImageNet class that is predicted by the pre-trained VGG-16 model. The output should always be an integer between 0 and 999, inclusive.

Before writing the function, make sure that you take the time to learn how to appropriately pre-process tensors for pre-trained models in the PyTorch documentation.

In [8]:
from PIL import Image
import torchvision.transforms as transforms

def VGG16_predict(img_path):
    '''
    Use pre-trained VGG-16 model to obtain index corresponding to 
    predicted ImageNet class for image at specified path
    
    Args:
        img_path: path to an image
        
    Returns:
        Index corresponding to VGG-16 model's prediction
    '''
    
    ## TODO: Complete the function.
    ## Load and pre-process an image from the given img_path
    ## Return the *index* of the predicted class for that image
    
    img = Image.open(img_path)
    
    transform = transforms.Compose([
                transforms.Resize(256),
                transforms.CenterCrop(224),
                transforms.ToTensor(),
                transforms.Normalize(mean=[0.485, 0.456, 0.406],
                                     std=[0.229, 0.224, 0.225])])
    
    # change model to eval mode
    VGG16.eval()
    img_t = transform(img)
    
    if use_cuda:
        img_t = img_t.cuda()
    
    img_t = torch.unsqueeze(img_t, 0)

    out = VGG16(img_t)
    
    _, ClassIndex = torch.max(out, 1)
    
    return ClassIndex.item() # predicted class index
In [9]:
im = Image.open(dog_files[100])
plt.imshow(im)

VGG16_predict(dog_files[100])
Out[9]:
170

(IMPLEMENTATION) Write a Dog Detector

While looking at the dictionary, you will notice that the categories corresponding to dogs appear in an uninterrupted sequence and correspond to dictionary keys 151-268, inclusive, to include all categories from 'Chihuahua' to 'Mexican hairless'. Thus, in order to check to see if an image is predicted to contain a dog by the pre-trained VGG-16 model, we need only check if the pre-trained model predicts an index between 151 and 268 (inclusive).

Use these ideas to complete the dog_detector function below, which returns True if a dog is detected in an image (and False if not).

In [10]:
### returns "True" if a dog is detected in the image stored at img_path
def dog_detector(img_path):
    ## TODO: Complete the function.
    
    index = VGG16_predict(img_path)
    
    if index >= 151 and index <= 268:
        return True
    
    return False # true/false

(IMPLEMENTATION) Assess the Dog Detector

Question 2: Use the code cell below to test the performance of your dog_detector function.

  • What percentage of the images in human_files_short have a detected dog?
  • What percentage of the images in dog_files_short have a detected dog?

Answer:

In [11]:
### TODO: Test the performance of the dog_detector function
### on the images in human_files_short and dog_files_short.
human = 0
dog = 0
for i in tqdm_notebook(range(100), desc='Detecting dogs...'):
    if dog_detector(human_files_short[i]) == True:
        human += 1
    if dog_detector(dog_files_short[i]) == True:
        dog += 1
        
print('Human face percentage {}%'.format(human))
print('Dog face percentage {}%'.format(dog))
Human face percentage 1%
Dog face percentage 97%

We suggest VGG-16 as a potential network to detect dog images in your algorithm, but you are free to explore other pre-trained networks (such as Inception-v3, ResNet-50, etc). Please use the code cell below to test other pre-trained PyTorch models. If you decide to pursue this optional task, report performance on human_files_short and dog_files_short.

In [12]:
### (Optional) 
### TODO: Report the performance of another pre-trained network.
### Feel free to use as many code cells as needed.

Res_Net50 = models.resnet50(pretrained=True)

if use_cuda:
    Res_Net50.cuda()
    
def dog_predict_resnet(img_path):
    img = Image.open(img_path)
    
    transform = transforms.Compose([
                transforms.Resize(256),
                transforms.CenterCrop(224),
                transforms.ToTensor(),
                transforms.Normalize(mean=[0.485, 0.456, 0.406],
                                     std=[0.229, 0.224, 0.225])])
    
    # change model to eval mode
    Res_Net50.eval()
    img_t = transform(img)
    
    if use_cuda:
        img_t = img_t.cuda()
        
    img_t = torch.unsqueeze(img_t, 0)

    out = Res_Net50(img_t)
    
    _, ClassIndex = torch.max(out, 1)
    
    return ClassIndex.item() # predicted class index


def dog_detector_resnet(img_path):
    ## TODO: Complete the function.
    
    index = dog_predict_resnet(img_path)
    
    if index >= 151 and index <= 268:
        return True
    
    return False # true/false

human = 0
dog = 0
for i in tqdm_notebook(range(100), desc='Detecting dogs...'):
    if dog_detector_resnet(human_files_short[i]) == True:
        human += 1
    if dog_detector_resnet(dog_files_short[i]) == True:
        dog += 1
        
print('Human face percentage {}%'.format(human))
print('Dog face percentage {}%'.format(dog))
Human face percentage 0%
Dog face percentage 99%

Step 3: Create a CNN to Classify Dog Breeds (from Scratch)

Now that we have functions for detecting humans and dogs in images, we need a way to predict breed from images. In this step, you will create a CNN that classifies dog breeds. You must create your CNN from scratch (so, you can't use transfer learning yet!), and you must attain a test accuracy of at least 10%. In Step 4 of this notebook, you will have the opportunity to use transfer learning to create a CNN that attains greatly improved accuracy.

We mention that the task of assigning breed to dogs from images is considered exceptionally challenging. To see why, consider that even a human would have trouble distinguishing between a Brittany and a Welsh Springer Spaniel.

Brittany Welsh Springer Spaniel

It is not difficult to find other dog breed pairs with minimal inter-class variation (for instance, Curly-Coated Retrievers and American Water Spaniels).

Curly-Coated Retriever American Water Spaniel

Likewise, recall that labradors come in yellow, chocolate, and black. Your vision-based algorithm will have to conquer this high intra-class variation to determine how to classify all of these different shades as the same breed.

Yellow Labrador Chocolate Labrador Black Labrador

We also mention that random chance presents an exceptionally low bar: setting aside the fact that the classes are slightly imabalanced, a random guess will provide a correct answer roughly 1 in 133 times, which corresponds to an accuracy of less than 1%.

Remember that the practice is far ahead of the theory in deep learning. Experiment with many different architectures, and trust your intuition. And, of course, have fun!

(IMPLEMENTATION) Specify Data Loaders for the Dog Dataset

Use the code cell below to write three separate data loaders for the training, validation, and test datasets of dog images (located at dog_images/train, dog_images/valid, and dog_images/test, respectively). You may find this documentation on custom datasets to be a useful resource. If you are interested in augmenting your training and/or validation data, check out the wide variety of transforms!

In [13]:
import os
import torch
from torchvision import datasets
import torchvision.transforms as transforms
import numpy as np

# check if CUDA is available
use_cuda = torch.cuda.is_available()

### TODO: Write data loaders for training, validation, and test sets
## Specify appropriate transforms, and batch_sizes

data_dir = 'data/dog_images/'
num_workers = 8

# transforms
train_transform = transforms.Compose([transforms.Resize(256),
                                      transforms.RandomRotation(30),
                                      transforms.RandomHorizontalFlip(),
                                      transforms.RandomResizedCrop(224),
                                      transforms.ToTensor(),
                                      transforms.Normalize((.5, .5, .5),
                                                           (.5, .5, .5))])

test_transform = transforms.Compose([transforms.Resize(256),
                                     transforms.CenterCrop(224),
                                     transforms.ToTensor(),
                                     transforms.Normalize((.5, .5, .5),
                                                          (.5, .5, .5))])


train_data = datasets.ImageFolder(data_dir + 'train', train_transform)
test_data = datasets.ImageFolder(data_dir + 'test', test_transform)
valid_data = datasets.ImageFolder(data_dir + 'valid', test_transform)

train_loader = torch.utils.data.DataLoader(train_data, batch_size=64, shuffle=True, num_workers=num_workers, pin_memory=True)
test_loader = torch.utils.data.DataLoader(test_data, batch_size=64, num_workers=num_workers, pin_memory=True)
valid_loader = torch.utils.data.DataLoader(valid_data, batch_size=64, num_workers=num_workers, pin_memory=True)

Question 3: Describe your chosen procedure for preprocessing the data.

  • How does your code resize the images (by cropping, stretching, etc)? What size did you pick for the input tensor, and why?
  • Did you decide to augment the dataset? If so, how (through translations, flips, rotations, etc)? If not, why not?

Answer:

  • I am resizing image to 256 px and then cropping it from center to 224 px. I am using this size for better performance of model. If size is high it will take more time. If I lower it further then image will lose valuable information.

  • Yes, I am rotaing image, flipping it horizontally and cropping it randomly.

(IMPLEMENTATION) Model Architecture

Create a CNN to classify dog breed. Use the template in the code cell below.

In [14]:
import torch.nn as nn
import torch.nn.functional as F

# define the CNN architecture
class Net(nn.Module):
    ### TODO: choose an architecture, and complete the class
    def __init__(self):
        super(Net, self).__init__()
        
        ## Define layers of a CNN        
        self.conv1 = nn.Conv2d(3, 32, 3, stride=1, padding=1)
        self.batch1 = nn.BatchNorm2d(32, affine=True, track_running_stats=True)
                
        self.conv2 = nn.Conv2d(32, 32, 3, stride=1, padding=1)
        self.batch2 = nn.BatchNorm2d(32, affine=True, track_running_stats=True)
                
        self.conv3 = nn.Conv2d(32, 64, 3, stride=1, padding=1)
        self.batch3 = nn.BatchNorm2d(64, affine=True, track_running_stats=True)        
        
        self.conv4 = nn.Conv2d(64, 64, 3, stride=1, padding=1)
        self.batch4 = nn.BatchNorm2d(64, affine=True, track_running_stats=True)
        
        self.conv5 = nn.Conv2d(64, 128, 3, stride=1, padding=1)
        self.batch5 = nn.BatchNorm2d(128, affine=True, track_running_stats=True)
        
        # half the output image
        self.pool = nn.MaxPool2d(2, 2)
        
        # conv layer output
        # final conv out 7 * 7 * 128
        
        self.fc1 = nn.Linear(7 * 7 * 128, 4096) 
        
        self.fc2 = nn.Linear(4096, 512)
        self.fcbn1 = nn.BatchNorm1d(num_features=512)
        
        self.fc3 = nn.Linear(512, 256)
        self.fcbn2 = nn.BatchNorm1d(num_features=256)
        
        self.fc4 = nn.Linear(256, 133)                
        
        self.drop = nn.Dropout2d(p=.3)
    
    def forward(self, x):
        ## Define forward behavior
        
        x = self.pool(self.batch1(F.relu(self.conv1(x), inplace=True)))
        x = self.pool(self.batch2(F.relu(self.conv2(x), inplace=True)))
        x = self.pool(self.batch3(F.relu(self.conv3(x), inplace=True)))        
        x = self.pool(self.batch4(F.relu(self.conv4(x), inplace=True)))                
        x = self.drop(x)
        x = self.pool(self.batch5(F.relu(self.conv5(x), inplace=True)))                
        x = self.drop(x)
        
        x = x.view(-1, 7 * 7 * 128)        
        
        x = F.relu(self.fc1(x))
        x = self.drop(x)
        x = self.fcbn1(F.relu(self.fc2(x)))
        x = self.drop(x)
        x = self.fcbn2(F.relu(self.fc3(x)))
        x = self.fc4(x)
        
        return x

#-#-# You so NOT have to modify the code below this line. #-#-#

# instantiate the CNN
model_scratch = Net()

# move tensors to GPU if CUDA is available
if use_cuda:
    model_scratch.cuda()
In [15]:
model_scratch
Out[15]:
Net(
  (conv1): Conv2d(3, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (batch1): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
  (conv2): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (batch2): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
  (conv3): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (batch3): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
  (conv4): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (batch4): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
  (conv5): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (batch5): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
  (pool): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  (fc1): Linear(in_features=6272, out_features=4096, bias=True)
  (fc2): Linear(in_features=4096, out_features=512, bias=True)
  (fcbn1): BatchNorm1d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
  (fc3): Linear(in_features=512, out_features=256, bias=True)
  (fcbn2): BatchNorm1d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
  (fc4): Linear(in_features=256, out_features=133, bias=True)
  (drop): Dropout2d(p=0.3, inplace=False)
)

Question 4: Outline the steps you took to get to your final CNN architecture and your reasoning at each step.

Answer:

First iteration

  • I started with small filter size like 8, 16, 32 with 3-4 convolution layers and 3 linear layers. Using this architecture when train loss reaches till 2 model started overfitting.

  • I used max pool layer of size (2,2) as standard. It make output of conv layer half in hieght and width.

  • I used kernel size 3 with padding 1

Second iteration

  • I realized that current filter count is not suitable for dog detection, as the dog is bigger object in image. So I increased filter count starting with 32, 64, 128 and 4 linear layer. Model this time generalizing well. I could achieve accuracy of 30% using this architecutre. I increased epoch to 150 to see if it can learn more, but to my disappointment it didnt generalise well.

  • No change in max pool and kernel sizes

  • Added dropout layer initially with .2 probablity and later increased it to .3.

Third and final iteration

  • Further research lead me to normalize input to hidden layers. As we are already normalizing input layer using torch normalize. I added BatchNorm2d after each conv layer. To my surprise, model generalize very well and it surpassed performance of last generation model. I trained it for 100 epoch and achieved accuracy of 40%.

  • Added one more conv layer

  • No change in max pool and kernel sizes

  • Added dropout in conv layer as well.

(IMPLEMENTATION) Specify Loss Function and Optimizer

Use the next code cell to specify a loss function and optimizer. Save the chosen loss function as criterion_scratch, and the optimizer as optimizer_scratch below.

In [16]:
import torch.optim as optim

### TODO: select loss function
criterion_scratch = nn.CrossEntropyLoss()

### TODO: select optimizer
optimizer_scratch = optim.Adadelta(model_scratch.parameters())

(IMPLEMENTATION) Train and Validate the Model

Train and validate your model in the code cell below. Save the final model parameters at filepath 'model_scratch.pt'.

In [17]:
loaders_scratch = {}
loaders_scratch['train'] = train_loader
loaders_scratch['test'] = test_loader
loaders_scratch['valid'] = valid_loader
In [18]:
from PIL import ImageFile
ImageFile.LOAD_TRUNCATED_IMAGES = True

train_losses = []
valid_losses = []

def train(n_epochs, loaders, model, optimizer, criterion, use_cuda, save_path):
    """returns trained model"""
    # initialize tracker for minimum validation loss
    valid_loss_min = np.Inf 
    
    for epoch in range(1, n_epochs+1):
        # initialize variables to monitor training and validation loss
        train_loss = 0.0
        valid_loss = 0.0
        
        ###################
        # train the model #
        ###################
        model.train()
        for batch_idx, (data, target) in enumerate(loaders['train']):
            # move to GPU
            if use_cuda:
                data, target = data.cuda(), target.cuda()
            ## find the loss and update the model parameters accordingly
            ## record the average training loss, using something like
            ## train_loss = train_loss + ((1 / (batch_idx + 1)) * (loss.data - train_loss))
            
            optimizer.zero_grad()
            
            output = model(data)
            
            loss = criterion(output, target)
            
            train_loss = train_loss + ( (1 / (batch_idx + 1)) * (loss.item() - train_loss) )
            
            loss.backward()
            optimizer.step()
            
        ######################    
        # validate the model #
        ######################
        model.eval()
        for batch_idx, (data, target) in enumerate(loaders['valid']):
            # move to GPU
            if use_cuda:
                data, target = data.cuda(), target.cuda()
            ## update the average validation loss
            
            output = model(data)
            loss = criterion(output, target)
            valid_loss = valid_loss + ( (1 / (batch_idx + 1)) * (loss.item() - valid_loss) )
            
        
        train_losses.append(train_loss)
        valid_losses.append(valid_loss)
        
        # print training/validation statistics 
        print('Epoch: {} \tTraining Loss: {:.6f} \tValidation Loss: {:.6f}'.format(
            epoch, 
            train_loss,
            valid_loss
            ))
        
        ## TODO: save the model if validation loss has decreased
        if valid_loss <= valid_loss_min:
            print('Validation loss decreased ({:.6f} --> {:.6f}). Saving model...'.format(valid_loss_min, valid_loss))
            valid_loss_min = valid_loss
            torch.save(model.state_dict(), save_path)
            
    # return trained model
    return model


# train the model
model_scratch = train(100, loaders_scratch, model_scratch, optimizer_scratch, 
                      criterion_scratch, use_cuda, 'model_scratch.pt')

# load the model that got the best validation accuracy
model_scratch.load_state_dict(torch.load('model_scratch.pt'))
Epoch: 1 	Training Loss: 4.831388 	Validation Loss: 4.693709
Validation loss decreased (inf --> 4.693709). Saving model...
Epoch: 2 	Training Loss: 4.626215 	Validation Loss: 4.665402
Validation loss decreased (4.693709 --> 4.665402). Saving model...
Epoch: 3 	Training Loss: 4.501663 	Validation Loss: 4.414899
Validation loss decreased (4.665402 --> 4.414899). Saving model...
Epoch: 4 	Training Loss: 4.420566 	Validation Loss: 4.375147
Validation loss decreased (4.414899 --> 4.375147). Saving model...
Epoch: 5 	Training Loss: 4.354198 	Validation Loss: 4.272185
Validation loss decreased (4.375147 --> 4.272185). Saving model...
Epoch: 6 	Training Loss: 4.298485 	Validation Loss: 4.351098
Epoch: 7 	Training Loss: 4.240793 	Validation Loss: 4.197996
Validation loss decreased (4.272185 --> 4.197996). Saving model...
Epoch: 8 	Training Loss: 4.168255 	Validation Loss: 4.058498
Validation loss decreased (4.197996 --> 4.058498). Saving model...
Epoch: 9 	Training Loss: 4.135219 	Validation Loss: 4.105325
Epoch: 10 	Training Loss: 4.082574 	Validation Loss: 4.042661
Validation loss decreased (4.058498 --> 4.042661). Saving model...
Epoch: 11 	Training Loss: 4.038181 	Validation Loss: 4.136009
Epoch: 12 	Training Loss: 3.995128 	Validation Loss: 4.001416
Validation loss decreased (4.042661 --> 4.001416). Saving model...
Epoch: 13 	Training Loss: 3.962981 	Validation Loss: 4.206322
Epoch: 14 	Training Loss: 3.919203 	Validation Loss: 4.178022
Epoch: 15 	Training Loss: 3.868634 	Validation Loss: 4.062803
Epoch: 16 	Training Loss: 3.842859 	Validation Loss: 3.857483
Validation loss decreased (4.001416 --> 3.857483). Saving model...
Epoch: 17 	Training Loss: 3.788532 	Validation Loss: 3.723223
Validation loss decreased (3.857483 --> 3.723223). Saving model...
Epoch: 18 	Training Loss: 3.768979 	Validation Loss: 3.741757
Epoch: 19 	Training Loss: 3.712341 	Validation Loss: 3.761645
Epoch: 20 	Training Loss: 3.679924 	Validation Loss: 3.729250
Epoch: 21 	Training Loss: 3.648231 	Validation Loss: 3.759917
Epoch: 22 	Training Loss: 3.615457 	Validation Loss: 3.474870
Validation loss decreased (3.723223 --> 3.474870). Saving model...
Epoch: 23 	Training Loss: 3.592137 	Validation Loss: 4.123848
Epoch: 24 	Training Loss: 3.568680 	Validation Loss: 3.753855
Epoch: 25 	Training Loss: 3.546885 	Validation Loss: 3.437114
Validation loss decreased (3.474870 --> 3.437114). Saving model...
Epoch: 26 	Training Loss: 3.508765 	Validation Loss: 3.449079
Epoch: 27 	Training Loss: 3.482886 	Validation Loss: 3.228741
Validation loss decreased (3.437114 --> 3.228741). Saving model...
Epoch: 28 	Training Loss: 3.442630 	Validation Loss: 3.439363
Epoch: 29 	Training Loss: 3.422835 	Validation Loss: 3.190033
Validation loss decreased (3.228741 --> 3.190033). Saving model...
Epoch: 30 	Training Loss: 3.409321 	Validation Loss: 3.522704
Epoch: 31 	Training Loss: 3.384916 	Validation Loss: 3.288482
Epoch: 32 	Training Loss: 3.363618 	Validation Loss: 3.385246
Epoch: 33 	Training Loss: 3.385330 	Validation Loss: 3.089876
Validation loss decreased (3.190033 --> 3.089876). Saving model...
Epoch: 34 	Training Loss: 3.344673 	Validation Loss: 3.237726
Epoch: 35 	Training Loss: 3.300379 	Validation Loss: 3.160512
Epoch: 36 	Training Loss: 3.321176 	Validation Loss: 3.216994
Epoch: 37 	Training Loss: 3.271213 	Validation Loss: 3.178918
Epoch: 38 	Training Loss: 3.299510 	Validation Loss: 3.341617
Epoch: 39 	Training Loss: 3.231529 	Validation Loss: 3.062702
Validation loss decreased (3.089876 --> 3.062702). Saving model...
Epoch: 40 	Training Loss: 3.228882 	Validation Loss: 2.936093
Validation loss decreased (3.062702 --> 2.936093). Saving model...
Epoch: 41 	Training Loss: 3.209699 	Validation Loss: 3.000313
Epoch: 42 	Training Loss: 3.208536 	Validation Loss: 3.062949
Epoch: 43 	Training Loss: 3.182697 	Validation Loss: 3.226667
Epoch: 44 	Training Loss: 3.185165 	Validation Loss: 3.066561
Epoch: 45 	Training Loss: 3.174851 	Validation Loss: 2.788487
Validation loss decreased (2.936093 --> 2.788487). Saving model...
Epoch: 46 	Training Loss: 3.149704 	Validation Loss: 2.925193
Epoch: 47 	Training Loss: 3.146251 	Validation Loss: 2.935817
Epoch: 48 	Training Loss: 3.085248 	Validation Loss: 2.994983
Epoch: 49 	Training Loss: 3.131578 	Validation Loss: 3.078694
Epoch: 50 	Training Loss: 3.097466 	Validation Loss: 3.018650
Epoch: 51 	Training Loss: 3.066013 	Validation Loss: 2.823811
Epoch: 52 	Training Loss: 3.091325 	Validation Loss: 3.245829
Epoch: 53 	Training Loss: 3.055194 	Validation Loss: 2.850901
Epoch: 54 	Training Loss: 3.040628 	Validation Loss: 3.008072
Epoch: 55 	Training Loss: 3.048293 	Validation Loss: 3.016021
Epoch: 56 	Training Loss: 3.022932 	Validation Loss: 2.772535
Validation loss decreased (2.788487 --> 2.772535). Saving model...
Epoch: 57 	Training Loss: 2.994931 	Validation Loss: 2.734986
Validation loss decreased (2.772535 --> 2.734986). Saving model...
Epoch: 58 	Training Loss: 3.018290 	Validation Loss: 2.699391
Validation loss decreased (2.734986 --> 2.699391). Saving model...
Epoch: 59 	Training Loss: 2.966159 	Validation Loss: 2.793090
Epoch: 60 	Training Loss: 2.948331 	Validation Loss: 2.927395
Epoch: 61 	Training Loss: 2.990239 	Validation Loss: 2.688802
Validation loss decreased (2.699391 --> 2.688802). Saving model...
Epoch: 62 	Training Loss: 2.953717 	Validation Loss: 2.712164
Epoch: 63 	Training Loss: 2.938993 	Validation Loss: 2.686833
Validation loss decreased (2.688802 --> 2.686833). Saving model...
Epoch: 64 	Training Loss: 2.919750 	Validation Loss: 2.614070
Validation loss decreased (2.686833 --> 2.614070). Saving model...
Epoch: 65 	Training Loss: 2.930621 	Validation Loss: 2.806738
Epoch: 66 	Training Loss: 2.918937 	Validation Loss: 2.770327
Epoch: 67 	Training Loss: 2.910790 	Validation Loss: 2.781004
Epoch: 68 	Training Loss: 2.903183 	Validation Loss: 2.648182
Epoch: 69 	Training Loss: 2.913112 	Validation Loss: 2.662931
Epoch: 70 	Training Loss: 2.884576 	Validation Loss: 2.720367
Epoch: 71 	Training Loss: 2.864133 	Validation Loss: 2.776986
Epoch: 72 	Training Loss: 2.878596 	Validation Loss: 2.631349
Epoch: 73 	Training Loss: 2.835336 	Validation Loss: 2.447746
Validation loss decreased (2.614070 --> 2.447746). Saving model...
Epoch: 74 	Training Loss: 2.863454 	Validation Loss: 2.471536
Epoch: 75 	Training Loss: 2.849097 	Validation Loss: 2.553757
Epoch: 76 	Training Loss: 2.826432 	Validation Loss: 2.461065
Epoch: 77 	Training Loss: 2.821166 	Validation Loss: 2.651972
Epoch: 78 	Training Loss: 2.823119 	Validation Loss: 2.357254
Validation loss decreased (2.447746 --> 2.357254). Saving model...
Epoch: 79 	Training Loss: 2.818573 	Validation Loss: 2.466153
Epoch: 80 	Training Loss: 2.811129 	Validation Loss: 2.449995
Epoch: 81 	Training Loss: 2.789730 	Validation Loss: 2.361486
Epoch: 82 	Training Loss: 2.776237 	Validation Loss: 2.488894
Epoch: 83 	Training Loss: 2.762761 	Validation Loss: 2.580532
Epoch: 84 	Training Loss: 2.770871 	Validation Loss: 2.412114
Epoch: 85 	Training Loss: 2.782292 	Validation Loss: 2.277864
Validation loss decreased (2.357254 --> 2.277864). Saving model...
Epoch: 86 	Training Loss: 2.718877 	Validation Loss: 2.511535
Epoch: 87 	Training Loss: 2.740583 	Validation Loss: 2.226871
Validation loss decreased (2.277864 --> 2.226871). Saving model...
Epoch: 88 	Training Loss: 2.762415 	Validation Loss: 2.366201
Epoch: 89 	Training Loss: 2.725870 	Validation Loss: 2.408538
Epoch: 90 	Training Loss: 2.679626 	Validation Loss: 2.437887
Epoch: 91 	Training Loss: 2.700047 	Validation Loss: 2.340524
Epoch: 92 	Training Loss: 2.703311 	Validation Loss: 2.319862
Epoch: 93 	Training Loss: 2.680081 	Validation Loss: 2.303363
Epoch: 94 	Training Loss: 2.674165 	Validation Loss: 2.232808
Epoch: 95 	Training Loss: 2.650028 	Validation Loss: 2.222461
Validation loss decreased (2.226871 --> 2.222461). Saving model...
Epoch: 96 	Training Loss: 2.680472 	Validation Loss: 2.197011
Validation loss decreased (2.222461 --> 2.197011). Saving model...
Epoch: 97 	Training Loss: 2.657748 	Validation Loss: 2.345035
Epoch: 98 	Training Loss: 2.645661 	Validation Loss: 2.319543
Epoch: 99 	Training Loss: 2.689250 	Validation Loss: 2.276079
Epoch: 100 	Training Loss: 2.644464 	Validation Loss: 2.291486
Out[18]:
<All keys matched successfully>
In [19]:
def plot_loss_curve(train_losses, valid_losses, nepoch):
    plt.figure(figsize=(15, 10))
    plt.plot(range(1, nepoch + 1), train_losses, label='Train')
    plt.plot(range(1, nepoch + 1), valid_losses, label='Valid')
    #plt.xticks(range(1, nepoch + 1), rotation=70)
    plt.xlabel('Epoch')
    plt.ylabel('Loss')
    plt.title('Loss curve')
    plt.legend();
In [20]:
plot_loss_curve(train_losses, valid_losses, 100)

(IMPLEMENTATION) Test the Model

Try out your model on the test dataset of dog images. Use the code cell below to calculate and print the test loss and accuracy. Ensure that your test accuracy is greater than 10%.

In [21]:
def test(loaders, model, criterion, use_cuda):

    # monitor test loss and accuracy
    test_loss = 0.
    correct = 0.
    total = 0.

    model.eval()
    for batch_idx, (data, target) in enumerate(loaders['test']):
        # move to GPU
        if use_cuda:
            data, target = data.cuda(), target.cuda()
        # forward pass: compute predicted outputs by passing inputs to the model
        output = model(data)
        # calculate the loss
        loss = criterion(output, target)
        # update average test loss 
        test_loss = test_loss + ((1 / (batch_idx + 1)) * (loss.data - test_loss))
        # convert output probabilities to predicted class
        pred = output.data.max(1, keepdim=True)[1]
        # compare predictions to true label
        correct += np.sum(np.squeeze(pred.eq(target.data.view_as(pred))).cpu().numpy())
        total += data.size(0)
            
    print('Test Loss: {:.6f}\n'.format(test_loss))

    print('\nTest Accuracy: %2d%% (%2d/%2d)' % (
        100. * correct / total, correct, total))
In [22]:
# call test function    
test(loaders_scratch, model_scratch, criterion_scratch, use_cuda)
Test Loss: 2.117357


Test Accuracy: 41% (345/836)

Step 4: Create a CNN to Classify Dog Breeds (using Transfer Learning)

You will now use transfer learning to create a CNN that can identify dog breed from images. Your CNN must attain at least 60% accuracy on the test set.

(IMPLEMENTATION) Specify Data Loaders for the Dog Dataset

Use the code cell below to write three separate data loaders for the training, validation, and test datasets of dog images (located at dogImages/train, dogImages/valid, and dogImages/test, respectively).

If you like, you are welcome to use the same data loaders from the previous step, when you created a CNN from scratch.

In [23]:
## TODO: Specify data loaders

(IMPLEMENTATION) Model Architecture

Use transfer learning to create a CNN to classify dog breed. Use the code cell below, and save your initialized model as the variable model_transfer.

In [24]:
import torchvision.models as models
import torch.nn as nn

## TODO: Specify model architecture 
model_transfer = models.vgg19_bn(pretrained=True)

# Freeze training for all "features" layers
for param in model_transfer.features.parameters():
    param.requires_grad = False

n_inputs = model_transfer.classifier[6].in_features

last_layer = nn.Linear(n_inputs, len(train_data.classes))

model_transfer.classifier[6] = last_layer


if use_cuda:
    model_transfer = model_transfer.cuda()

Question 5: Outline the steps you took to get to your final CNN architecture and your reasoning at each step. Describe why you think the architecture is suitable for the current problem.

Answer:

I am using VGG19 with batch normalization. VGG19 has 8.15 Top-5 error and it is small in size compared to other models. Since, we have 133 classes to predict, so I am changing last linear layer to match output.

(IMPLEMENTATION) Specify Loss Function and Optimizer

Use the next code cell to specify a loss function and optimizer. Save the chosen loss function as criterion_transfer, and the optimizer as optimizer_transfer below.

In [25]:
criterion_transfer = nn.CrossEntropyLoss()
optimizer_transfer = optim.Adadelta(model_transfer.parameters())

(IMPLEMENTATION) Train and Validate the Model

Train and validate your model in the code cell below. Save the final model parameters at filepath 'model_transfer.pt'.

In [26]:
# train the model
model_transfer = train(10, loaders_scratch, model_transfer, optimizer_transfer, criterion_transfer, use_cuda, 'model_transfer.pt')
Epoch: 1 	Training Loss: 2.793619 	Validation Loss: 1.486570
Validation loss decreased (inf --> 1.486570). Saving model...
Epoch: 2 	Training Loss: 2.162689 	Validation Loss: 1.394425
Validation loss decreased (1.486570 --> 1.394425). Saving model...
Epoch: 3 	Training Loss: 2.015152 	Validation Loss: 1.271365
Validation loss decreased (1.394425 --> 1.271365). Saving model...
Epoch: 4 	Training Loss: 1.960864 	Validation Loss: 1.038778
Validation loss decreased (1.271365 --> 1.038778). Saving model...
Epoch: 5 	Training Loss: 1.950669 	Validation Loss: 0.904052
Validation loss decreased (1.038778 --> 0.904052). Saving model...
Epoch: 6 	Training Loss: 1.919604 	Validation Loss: 1.141878
Epoch: 7 	Training Loss: 1.887840 	Validation Loss: 1.000381
Epoch: 8 	Training Loss: 1.841412 	Validation Loss: 0.753305
Validation loss decreased (0.904052 --> 0.753305). Saving model...
Epoch: 9 	Training Loss: 1.853138 	Validation Loss: 1.163234
Epoch: 10 	Training Loss: 1.867027 	Validation Loss: 0.923917
In [27]:
# load the model that got the best validation accuracy (uncomment the line below)
model_transfer.load_state_dict(torch.load('model_transfer.pt'))
Out[27]:
<All keys matched successfully>

(IMPLEMENTATION) Test the Model

Try out your model on the test dataset of dog images. Use the code cell below to calculate and print the test loss and accuracy. Ensure that your test accuracy is greater than 60%.

In [28]:
test(loaders_scratch, model_transfer, criterion_transfer, use_cuda)
Test Loss: 0.829069


Test Accuracy: 75% (635/836)

(IMPLEMENTATION) Predict Dog Breed with the Model

Write a function that takes an image path as input and returns the dog breed (Affenpinscher, Afghan hound, etc) that is predicted by your model.

In [29]:
### TODO: Write a function that takes a path to an image as input
### and returns the dog breed that is predicted by the model.

# list of class names by index, i.e. a name can be accessed like class_names[0]
class_names = [item[4:].replace("_", " ") for item in train_data.classes]

def predict_breed_transfer(img_path):
    # load the image and return the predicted breed
    img = Image.open(img_path)
    
    # change model to eval mode
    model_transfer.eval()
    img_t = test_transform(img)
    
    if use_cuda:
        img_t = img_t.cuda()
    
    img_t = torch.unsqueeze(img_t, 0)

    out = model_transfer(img_t)
    
    _, ClassIndex = torch.max(out, 1)
    
    return class_names[ClassIndex.item()] # predicted class index
    
In [30]:
im = Image.open(dog_files[90])
plt.imshow(im)

predict_breed_transfer(dog_files[90])
Out[30]:
'Tibetan mastiff'

Step 5: Write your Algorithm

Write an algorithm that accepts a file path to an image and first determines whether the image contains a human, dog, or neither. Then,

  • if a dog is detected in the image, return the predicted breed.
  • if a human is detected in the image, return the resembling dog breed.
  • if neither is detected in the image, provide output that indicates an error.

You are welcome to write your own functions for detecting humans and dogs in images, but feel free to use the face_detector and human_detector functions developed above. You are required to use your CNN from Step 4 to predict dog breed.

Some sample output for our algorithm is provided below, but feel free to design your own user experience!

Sample Human Output

(IMPLEMENTATION) Write your Algorithm

In [43]:
### TODO: Write your algorithm.
### Feel free to use as many code cells as needed.

def run_app(img_path):
    ## handle cases for a human face, dog, and neither    
    other = False
    
    message = ''
    
    # check human
    if face_detector(img_path) == True:
        print('Hello human!')
        message = 'You look like '
    elif dog_detector(img_path) == True:
        print('Hey, doggy!')
        message = 'You are '
    else:
        print('You are niether dog nor human :(')
        other = True
    
    plt.figure(figsize=(6,6))
    im = Image.open(img_path)
    plt.imshow(im)
    plt.show()
    
    if other == False:
        breed = predict_breed_transfer(img_path)
        print(message, breed)
        print()
In [32]:
run_app(dog_files[23])
Hey, doggy!
You are  Italian greyhound

In [33]:
run_app(dog_files[99])
Hey, doggy!
You are  Irish wolfhound


Step 6: Test Your Algorithm

In this section, you will take your new algorithm for a spin! What kind of dog does the algorithm think that you look like? If you have a dog, does it predict your dog's breed accurately? If you have a cat, does it mistakenly think that your cat is a dog?

(IMPLEMENTATION) Test Your Algorithm on Sample Images!

Test your algorithm at least six images on your computer. Feel free to use any images you like. Use at least two human and two dog images.

Question 6: Is the output better than you expected :) ? Or worse :( ? Provide at least three possible points of improvement for your algorithm.

Answer: (Three possible points for improvement)

  • The algorithm is working correctly. It is even detecting Anubis as not human not dog whereas anubis is dog human hybrid.
In [35]:
## TODO: Execute your algorithm from Step 6 on
## at least 6 images on your computer.
## Feel free to use as many code cells as needed.

## suggested code, below
#for file in np.hstack((human_files[:3], dog_files[:3])):
#    run_app(file)
    
run_app('./images/Me.jpeg')
Hello human!
You look like  Basset hound

In [47]:
run_app('./images/anubis.jpeg')
You are niether dog nor human :(
In [37]:
run_app('./images/DSC00730.JPG')
Hey, doggy!
You are  Golden retriever

In [39]:
run_app('./images/American_water_spaniel_00648.jpg')
Hey, doggy!
You are  Curly-coated retriever

In [44]:
run_app('./images/cat.jpeg')
You are niether dog nor human :(
In [45]:
run_app('./images/lion.jpeg')
You are niether dog nor human :(